Risk and Rationality in Adolescent Decision Making

Transcription

1 PSYCHOLOGICAL SCIENCE IN THE PUBLIC INTEREST Risk and Rationality in Adolescent Decision Making Implications for Theory, Practice, and Public Policy Valerie F. Reyna 1 and Frank Farley 2 1 Cornell University and 2 Temple University SUMMARY Crime, smoking, drug use, alcoholism, reckless driving, and many other unhealthy patterns of behavior that play out over a lifetime often debut during adolescence. Avoiding risks or buying time can set a different lifetime pattern. Changing unhealthy behaviors in adolescence would have a broad impact on society, reducing the burdens of disease, injury, human suffering, and associated economic costs. Any program designed to prevent or change such risky behaviors should be founded on a clear idea of what is normative (what behaviors, ideally, should the program foster?), descriptive (how are adolescents making decisions in the absence of the program?), and prescriptive (which practices can realistically move adolescent decisions closer to the normative ideal?). Normatively, decision processes should be evaluated for coherence (is the thinking process nonsensical, illogical, or self-contradictory?) and correspondence (are the outcomes of the decisions positive?). Behaviors that promote positive physical and mental health outcomes in modern society can be at odds with those selected for by evolution (e.g., early procreation). Healthy behaviors may also conflict with a decision maker s goals. Adolescents goals are more likely to maximize immediate pleasure, and strict decision analysis implies that many kinds of unhealthy behavior, such as drinking and drug use, could be deemed rational. However, based on data showing developmental changes in goals, it is important for policy to promote positive long-term outcomes rather than adolescents short-term goals. Developmental data also suggest that greater risk aversion is generally adaptive, and that decision processes that support this aversion are more advanced than those that support risk taking. Address correspondence to Valerie F. Reyna, Department of Human Development, B44 MVR Hall, Cornell University, Ithaca, NY 14853; vr53@cornell.edu. A key question is whether adolescents are developmentally competent to make decisions about risks. In principle, barring temptations with high rewards and individual differences that reduce self-control (i.e., under ideal conditions), adolescents are capable of rational decision making to achieve their goals. In practice, much depends on the particular situation in which a decision is made. In the heat of passion, in the presence of peers, on the spur of the moment, in unfamiliar situations, when trading off risks and benefits favors bad long-term outcomes, and when behavioral inhibition is required for good outcomes, adolescents are likely to reason more poorly than adults do. Brain maturation in adolescence is incomplete. Impulsivity, sensation seeking, thrill seeking, depression, and other individual differences also contribute to risk taking that resists standard risk-reduction interventions, although some conditions such as depression can be effectively treated with other approaches. Major explanatory models of risky decision making can be roughly divided into (a) those, including health-belief models and the theory of planned behavior, that adhere to a rational behavioral decision-making framework that stresses deliberate, quantitative trading off of risks and benefits; and (b) those that emphasize nondeliberative reaction to the perceived gists or prototypes in the immediate decision environment. (A gist is a fuzzy mental representation of the general meaning of information or experience; a prototype is a mental representation of a standard or typical example of a category.) Although perceived risks and especially benefits predict behavioral intentions and risk-taking behavior, behavioral willingness is an even better predictor of susceptibility to risk taking and has unique explanatory power because adolescents are willing to do riskier things than they either intend or expect to do. Dual-process models, such as the prototype/willingness model and fuzzy-trace theory, identify two divergent paths to risk taking: a reasoned and a reactive route. Such Volume 7 Number 1 Copyright r 2006 Association for Psychological Science 1

2 Risk and Rationality in Adolescent Decision Making models explain apparent contradictions in the literature, including different causes of risk taking for different individuals. Interventions to reduce risk taking must take into account the different causes of such behavior if they are to be effective. Longitudinal and experimental research are needed to disentangle opposing causal processes particularly, those that produce positive versus negative relations between risk perceptions and behaviors. Counterintuitive findings that must be accommodated by any adequate theory of risk taking include the following: (a) Despite conventional wisdom, adolescents do not perceive themselves to be invulnerable, and perceived vulnerability declines with increasing age; (b) although the object of many interventions is to enhance the accuracy of risk perceptions, adolescents typically overestimate important risks, such as HIV and lung cancer; (c) despite increasing competence in reasoning, some biases in judgment and decision making grow with age, producing more irrational violations of coherence among adults than among adolescents and younger children. The latter occurs because of a known developmental increase in gist processing with age. One implication of these findings is that traditional interventions stressing accurate risk perceptions are apt to be ineffective or backfire because young people already feel vulnerable and overestimate their risk. In addition, research shows that experience is not a good teacher for children and younger adolescents, because they tend to learn little from negative outcomes (favoring the use of effective deterrents, such as monitoring and supervision), although learning from experience improves considerably with age. Experience in the absence of negative consequences may increase feelings of invulnerability and thus explain the decrease in risk perceptions from early to late adolescence, as exploration increases. Finally, novel interventions that discourage deliberate weighing of risks and benefits by adolescents may ultimately prove more effective and enduring. Mature adults apparently resist taking risks not out of any conscious deliberation or choice, but because they intuitively grasp the gists of risky situations, retrieve appropriate risk-avoidant values, and never proceed down the slippery slope of actually contemplating tradeoffs between risks and benefits. INTRODUCTION In this monograph, we review scientific evidence concerning the causes and remediation of unhealthy risk taking in adolescence. Adolescent risk taking has economic, psychological, and health implications (e.g., Maynard, 1997). Smoking, drug use, unprotected sex, and unsafe driving take demonstrable tolls in healthcare costs and property damage, as well as less readily measured costs in human misery and lost potential. Habits begun at this age can last a lifetime. Table 1 shows one set of prevalence measures for adolescents. Opinions about proper solutions to the problem of unhealthy adolescent risk taking are plentiful, ranging from abstinence education to higher legal drinking ages. However, the public and policymakers rarely make use of the scientific literature on risky decision making in adolescence, and, as in many areas of human behavior, prevention and intervention programs are generally not based on such evidence. Those seeking a comprehensive view of the evidence (and not just the bits supporting one s own favored position) need to cast a wide net. One of the barriers to more comprehensive use of the scientific literature is the fragmentation of research. Relevant studies are scattered across disciplines (e.g., psychology, sociology, pediatrics, public health) and problem-specific professional communities (e.g., smoking, AIDS prevention, alcohol and substance abuse) whose members attend specialized conferences and read specialized journals, and who are sometimes isolated further by adherence to specific research paradigms or treatment modalities. To be sure, specialization is necessary if scholars are to apprehend the vast amount of research within particular problem domains. For example, the biochemistry of smoking and alcohol are each complex enough to justify separate expertise. The effects of alcohol on brain development and on psychomotor skills (e.g., driving) are themselves different enough to direct scholars and practitioners to separate conferences and publications. However, fragmentation exacts a price. Relevant work is published that escapes notice in closely related domains (e.g., smoking versus alcohol use) and explanatory models found useful in one domain are not necessarily considered in other domains. There is also the problem of reinventing the wheel. For example, Dawes and Corrigan (1974; Dawes, 1979) found that many competing models of decision-making processes were inherently indistinguishable because of their shared statistical properties. 1 Additionally, the commonalities among laboratory and real world tasks argued to reflect risk taking need to be identified and limits of commonality or generalizability established. Risk taking in a laboratory task involving minor symbolic risks may have little to do with the risk taking of a carload of drunk adolescents on the interstate on a Friday night (Farley, 1996). Hence, a cross-cutting analysis is urgently needed to identify the findings and explanatory models that generalize across domains, as well as the domain-specific limits to generalization. To address this need, we examine one topic that generalizes across domains: the optimality of adolescents decisions about 1 Attempts to reconcile the weights afforded to various factors in different studies were doomed to failure, because they reflected uninteresting measurement issues. In fact, Dawes s conclusion was an inductive rediscovery of principles derived deductively by Wilks (1937), in an even more general look at the properties of linear models. 2 Volume 7 Number 1

3 Valerie F. Reyna and Frank Farley TABLE 1 Percentages of Youth in 9th Through 12th Grades Engaging in Various Risk and Risk-Preventive Behaviors, From National Youth Risk Behavior Survey ( ) Behavior Data Change: Change: Seatbelts a Decrease None Drinking driver b Decrease None Carried weapon c Decrease None Fighting d Decrease None Suicide attempt e None None Cigarette use f Decrease Decrease Smokeless tobacco g NA NA Decrease None Alcohol use h Decrease None Marijuana use i Increase None Cocaine use j Increase None Sexual intercourse k Decrease None Condom use l Increase Increase Exercise m NA None None Physical education n Decrease None Overweight o NA NA NA NA None None Note. The data are from Centers for Disease Control and Prevention (2004). a Rarely wore seatbelts while riding in a car driven by someone else b Rode in a car with a driver who has been drinking alcohol during the past 30 days c Carried a gun, knife, or club at least once during the past 30 days d Was in a physical fight at least once during the past 12 months e Attempted suicide at least once during the past 12 months f Smoked cigarettes on at least 1 day during the past 30 days g Used chewing tobacco, snuff, or dip on at least 1 day during the past 30 days h Drank alcohol on at least 1 day during the past 30 days i Used marijuana on at least 1 day during the past 30 days j Used cocaine on at least 1 day during the past 30 days k Has had sexual intercourse at least once l Used condoms during sexual intercourse m Participated in vigorous physical exercise for at least 20 minutes on at least 3 of the past 7 days n Attends physical education class daily o Is above the 95th percentile for the body mass index, by age and sex norms risky behaviors. We consider both the processes involved and the performance levels that adolescents achieve and could achieve, with possible interventions. Knowing those levels is critical to creating sound policies concerning such issues as drinking age and adolescents culpability for crimes, informed consent for medical procedures, and responsiveness to AIDSprevention curricula. We recognize that adolescents choices reflect the interaction of general skills and specific situational demands, which together determine the bounds of rationality in adolescence. Thus, in this article, we discuss the mounting evidence about adolescent rationality and the implications of this evidence for problem behaviors. Owing to the voluminous and fragmented nature of the literature, our review is not the conventional sort in which every article fitting some set of inclusion criteria is examined, effect sizes are calculated, and a single question (say, about effectiveness of pregnancy-prevention curricula) is asked and answered. Although we undertook such a conventional review before writing this paper, to ensure that our judgments are firmly grounded in current work, space does not permit us to discuss or even to mention every scientific article on adolescent risk taking. Instead, our aim is to provide a solid, empirically grounded framework for understanding adolescent risk taking and determining what it would take to reduce or eliminate unhealthy behaviors. Many unanswered questions concerning the nature of adaptive behavior, healthy risk taking, and rational decision making in adolescence remain. Notwithstanding the limitations of current knowledge, however, scientists have learned a great deal that can be useful today. Extant data identify successful practices (e.g., effective curricula for reducing risk taking) and promising practices that have yet to be studied systematically. Existing data also demonstrate that some common beliefs, such as the belief that adolescents feel uniquely invulnerable, are myths. Questions that are addressed by current data include the following, which provide an outline for the remainder of this review: Why is adolescent risky decision making important? What is rational, adaptive, or good decision making for adolescents? What are the main explanatory models of adolescent risk taking? Volume 7 Number 1 3

4 Risk and Rationality in Adolescent Decision Making What are the key data more particularly, the data that illuminate prediction, explanation, and intervention? What are some key implications of current findings for different approaches to risk reduction and avoidance? BACKGROUND AND PERSPECTIVES The intended audience for this monograph is anyone who wants to become acquainted with current scientific evidence concerning the causes and remediation of unhealthy risk taking in adolescence, including those for whom policy, practice, or prevention is the main motivation. A narrated list of findings, however, would be insufficient to address this topic. The first and most fundamental question is how to know what unhealthy risk taking is. The answer may seem obvious, but noted scholars have disagreed vehemently about this issue. So before we examine the issue in depth, we give the reader a sense of why the answer is not obvious and how the answer shapes thinking about unhealthy risk taking and its remedies. We then explain why evidence-based theories of risky decision making cannot be ignored if we wish to understand and apply the findings regarding adolescent risky decision making to improve lives. In short, if the goal is to change behavior in a positive direction, it is crucial to know more than a list of findings about risky adolescent decision making: It is crucial to know what the desired endpoint ( positive change ) is and how to measure it, and to know which explanations of behavior are likely to be true, based on the evidence. Thus, we review specific theories of behavior change and decision making because, in our view, these theories offer the best account of the evidence to date. Theories, in this usage of that term, are summaries and explanations of evidence, not speculations or philosophical arguments. How can we know what unhealthy risk taking is? Although perspectives on how to tell if decision making is good or bad differ, each one captures important aspects of the data. Ultimately, we include both of the major schools of thought (coherence and correspondence) in our criteria for rational decision making, but others might justifiably side with one view rather than another (we present our arguments in depth later). Traditional theories of rational decision making indicate that either risk taking or risk aversion can be rational, as long as the decision process is coherent (i.e., internally consistent). Traditional decision-making theorists do not make judgments about what people believe, and they would characterize many of the behaviors that society might wish to discourage among adolescents as rational. Although some might disagree with these conclusions about rationality, traditional theories point up factors that have been shown to influence risk-taking behavior in adolescence and, if the theories are true, they identify which policies and practices are likely to be effective in reducing risk taking (although new theories, discussed below, suggest that reducing unhealthy risk taking requires more than rational reasoning skills). Traditional theories distinguish rational decision processes from good outcomes because outcomes are determined by many factors outside of the decision process. Someone cannot be described as engaging in unhealthy risk taking if there is no rational basis to predict that, for unforeseeable reasons, the outcome will turn out to be bad. Critics of traditional theories disagree that outcomes are irrelevant to judging the quality of decision making and, on the contrary, disparage coherence of decision processes as a criterion of rationality. In this correspondence view, good outcomes signal good decision making. Correspondence refers to correspondence to reality, which outcomes reflect. Although this view has superficial appeal, there are numerous documented examples of decision makers who enjoyed good outcomes by accident (having made clear mistakes in judgment) and vice versa. The adolescent who has unprotected sex numerous times without getting pregnant could argue, in this view, that her behavior is perfectly rational because she has avoided an undesirable outcome. Clearly, the correspondence view has shortcomings that are not apparent at first blush. Some evolutionary theorists have also criticized traditional coherence approaches to rationality, arguing that violations of logic or probability or other rules of coherence are apparent rather than real and that evolution gives human decision makers simple heuristics that make us smart (Gigerenzer, Todd, & the ABC Group, 1999). However, these simple gut-level decisions that are encouraged by evolution appear to make people stupid in the modern world under predictable circumstances, and they encourage unhealthy risk taking rather than discourage it. (Naturally, such behaviors may have been adaptive at an earlier point in evolutionary history.) The realm of adolescent decision making, therefore, provides a counterexample to the general claim made by some evolutionary theorists that the smart choices in one s work or personal life are those selected for by evolution. It is useful for prevention and intervention efforts to acknowledge that adolescents may have to resist evolutionary pressures that promote consuming substances that offer immediate pleasure or having sex before they are prepared for its economic and psychological consequences. We do not claim that evolutionary theories are irrelevant, and we cite several books for further reading in this area, such as those by Baumeister (2005); Geary (2005); and Gigerenzer, Todd, and the ABC Group (1999). Evolutionary theory, and the construct of adaptive behavior, is central to understanding rationality in the correspondence sense (i.e., which decision processes and behaviors promote positive long-term outcomes). However, evolutionary claims that are made on the basis of philosophical arguments, mathematical proofs not involving observables, and hypothetical computer simulations should be sharply distinguished from claims that have been tested empirically. If the policy recommendations of social scientists are to be taken seriously, it is necessary to retain scientific credibility by sticking to empirical evidence and to theories that 4 Volume 7 Number 1

5 Valerie F. Reyna and Frank Farley are grounded in empirical evidence. In order to be ready for consideration at the level of policy, promising evolutionary theories should be subjected to the same kinds of empirical tests as the core theories that we discuss below. Whichever view of rationality one takes (traditional coherence, correspondence, or, at some future point when more data are gathered, evolutionary), it is essential to consider the developmental differences between adolescents and adults when judging their behavior. The traditional coherence view emphasizes the centrality of making choices that allow the decision maker to reach his or her own goals. As we discuss, evidence on developmental differences raises the specter that goals change with age, and the issue is then which goals (adolescents current goals or their inferred future goals) to consider in judging rationality. If rationality also demands (as it must in traditional views) that decision processes be logical, then it also makes sense to ask whether adolescents are capable of thinking logically. We briefly review the data on that issue as well. Other developmental differences relevant to judging rationality, including impulsivity, are also reviewed. Laboratory data on developmental differences in probability judgment and in decision making for example, involving choosing between sure things and gambles are also relevant to the kinds of psychological competence that underlie risk-taking behavior. The developmental questions are: What do children (and, subsequently, adolescents) know and when do they know it? On the one hand, laboratory studies have shown that young children trade off the probability of winning a prize and the number of prizes to be won (essentially multiplying the odds of winning by the amount to be won, and choosing accordingly; e.g., Reyna & Ellis, 1994). On the other hand, analogous studies of probability judgment and choice in adults have been the source of numerous illustrations of cognitive illusions namely, adults ignoring objective information about probabilities and outcomes and instead basing their responses on illusory stereotypes or superficial wording of decision scenarios (e.g., Gilovich, Griffin, & Kahneman, 2002). However, this seeming contradiction between early analytic competence and late-persisting cognitive illusions can be explained by modern developmental theories (e.g., fuzzy-trace theory) that predict exactly these kinds of paradoxical patterns. The theories that we review are older ones that have amassed the most definitive evidence about causal factors in risky decision making, and newer ones that enjoy the advantage of building on the discoveries of the pre-existing models, thus being able to improve on their predictions. We should add that all of the models we review, including the older models, should be considered currently relevant and that, although the data favor newer models, those data are far from extensive at this point. Traditional models are those that essentially adhere to the behavioral decision framework, which would include such rational deliberative approaches as health-belief models, the theory of reasoned action, the theory of planned behavior, problem-solving approaches, and other similar theories including some with less evidence (that we, therefore, do not discuss). Concepts that figure in such models include perceived risks and benefits, social norms (beliefs about other people, such as whether one s parents approve of underage drinking or whether peers are engaging in sex), self-efficacy (beliefs about being competent in a specific domain or skill, such as being capable of standing up to pressure to have sex), perceived control, and behavioral intention. Newer models of adolescent risky decision making include the prototype/willingness model and fuzzy-trace theory. In these models, risk taking is determined by mental representations of risk takers (e.g., smokers) or risky situations (e.g., a couple alone in a hotel room on prom night), along with other factors such as willingness (as opposed to intention, in the prototype/willingness model) and situationdependent retrieval of risk-avoidant values (in fuzzy-trace theory). These traditional and newer models aim to describe and explain real behavior. However, they also typically incorporate assumptions about what constitutes ideal behavior, and thus provide a goal for prescriptive interventions to improve decision making. In each of these models, perception of risks plays an important role (although how people think about risk is construed very differently across models). There are different ways to assess risk perception that seem to yield different conclusions, but those conclusions are actually compatible. A concrete example may be helpful: Imagine an adolescent who has sex without a condom and who overestimates the risk of contracting a sexually transmitted disease but overestimates his own risk less than he does that of comparable others (e.g., other adolescents who have sex without condoms) an optimistic bias. Furthermore, imagine that this adolescent rates his own risk of getting a sexually transmitted disease as higher than adolescents who use a condom rate their own risk and as higher than adults rate their own risk (regardless of whether they use a condom or not). (Conditional assessments, such as estimating the risk of acquiring sexually transmitted diseases if one has sex without a condom, do not change the result that adolescents rate themselves as more vulnerable than adults rate themselves.) As is apparent from this example, these comparisons suggest different messages about perceived vulnerability if taken in isolation from one another, but they are not mutually exclusive. This adolescent overestimates the level of objective risk, displays an optimistic bias relative to others, and yet acknowledges that he is at higher risk than adults and adolescents not engaging in specific risk-taking behaviors. Based on the literature, we can say that this adolescent is typical, as these results tend to be found consistently (except with respect to comparisons between lowerrisk and higher-risk adolescents, which have produced variable results). To preview our later discussion, the key descriptive findings regarding adolescents perception of risks are these: Much like adults, most adolescents exhibit an optimistic bias, in which Volume 7 Number 1 5

6 Risk and Rationality in Adolescent Decision Making they view their own risks as less than those of comparable peers. However, objectively higher-risk groups sometimes estimate their risk as higher, and sometimes as lower, than lower-risk groups rate themselves. For example, Johnson, McCaul, and Klein (2002) found that adolescents who were daily smokers and those engaged in unprotected sex estimated their risk of getting lung cancer or a sexually transmitted disease, respectively, as significantly higher than others not engaging in those behaviors did. Some studies confirm this pattern; other studies report no difference or lower perceived risks among those engaging in risk-taking behavior. As we discuss, measures matter; how the question about risk is asked makes a difference (Fishbein, 2003). The role of experienced outcomes may also explain these variable findings (experiencing negative outcomes may increase risk estimates and failing to experience negative outcomes may do the opposite), but preliminary evidence on this point is meager. A consistent finding that emerges from this literature, and one that has been replicated in different laboratories, is that the optimistic bias is no more prevalent in adolescents than it is in adults, and, indeed, adolescents perceive themselves as more vulnerable than adults perceive themselves to be. In addition, when subjective and objective estimates of risk can be compared, adolescents tend to overestimate important risks (e.g., of HIV infection or lung cancer), although they may underestimate harmful consequences and long-term effects, such as addiction. They think that the risk is high, but the consequences are not that bad. (Not all risks are overestimated; unfamiliar risks that are not covered in health curricula, such as the risk of food poisoning, might well be underestimated.) Another consistent finding is that, when they are directly compared, benefits loom larger than risks. That is, perceived benefits predict risk-taking behavior and often carry more weight than perceived risks do. Thus, despite overestimation of risks, perceived benefits may drive adolescents reactive behaviors and behavioral intentions, explaining why adolescents who perceive risks to be high would still take those risks. Nevertheless, constructs such as perceived risks and benefits do not explain all risk taking for all adolescents; there is variance in risk-taking behaviors that is not accounted for by traditional models. The bottom line of the data concerning extant models is that the older models of deliberative decision making (resulting in behavioral intentions and planned behaviors) fail to account for a substantial amount of adolescent risk taking, which is spontaneous, reactive, and impulsive. This conclusion about gaps in older models holds even when higher methodological standards, such as conditional risk assessments (e.g., estimating the risk of acquiring sexually transmitted diseases if one has sex without a condom) and prospective designs that control for initial perceptions and behavior, are used in research (see Brewer, Weinstein, Cuite, & Herrington, 2004; Gerrard, Gibbons, Benthin, & Hessling, 1996; Weinstein & Nicolich, 1993; and Weinstein, Rothman, & Nicolich, 1998, for details concerning design and methodology). To be sure, methodological improvements have resulted in improved support for traditional models, yielding, for example, stronger and more consistent relations between perceived risk and behavior (e.g., Brewer et al., in press; Fishbein, 2003). However, almost all of the patterns of findings we discuss that pertain to adolescents remain robust despite these methodological modifications. We do not conclude that traditional models are worthless. On the contrary, there is ample evidence favoring such models and, simultaneously, evidence indicating that they have important gaps. We resolve this dilemma by acknowledging, based on the data, that adolescents apparently make decisions in different ways namely, deliberately, reactively, and intuitively. Deliberate decision making is explained by traditional models; reactive decision making is explained by the prototype/ willingness model; and fuzzy-trace theory explains intuitive decision making, contrasting gist-based intuition to avoid risk with deliberation that encourages risk taking. (A gist is a fuzzy mental representation of the general meaning of information or experience, and gist-based intuition is reasoning or decision making based on these fuzzy representations.) As we discuss in some detail, ideas about emotion (as temptation and as a healthy cue) and personal experience with risks are being increasingly incorporated into contemporary theories, including traditional behavioral decision-making approaches. We have mentioned decisions that result from deliberation, reaction, and intuition, but one might also ask about decisions that come about through imitation, habit, social conventions, and social heuristics (see Nisbett & Ross, 1980, for an excellent discussion of some of these classic issues). These social factors are reflected in perceived social norms, images or prototypes, perceived benefits, and other constructs that we have reviewed, and, thus, they are indirectly represented to some extent in the theoretical approaches that we discuss. For example, adolescents might follow a social heuristic to do what the majority does, which would be reflected in perceived social norms (beliefs about what the majority does) and perceived benefits (the belief that doing what the majority does ensures being accepted by one s peer group, a social benefit). One might imagine that responding to these social factors could be reactive (going along with the majority without thinking) or deliberative (calculating that one would pay too high a price socially by opposing the majority). Habits are also, again indirectly, related to decision making in standard dual-process theories, as they reflect the operation of an evolutionarily older, associative system (Kahneman, 2003; Sloman, 2002). However, although much evidence supports the effects of imitation, habits, and social conventions on behavior, the connections between these effects and constructs in decision theories of adolescent risk taking have yet to be fully elaborated. Therefore, we do not discuss these factors further, except to acknowledge that habits (and addictions) established in adolescence can perpetuate behaviors that older decision makers would not have initiated. We do 6 Volume 7 Number 1

7 Valerie F. Reyna and Frank Farley discuss the role of social payoffs and other rationalizations for adolescent risk taking in the context of theories of rationality. As we explain later, merely asserting that risk taking has social benefits (e.g., peer acceptance) does not necessarily justify such behavior nor does it prove that the behavior is rational. Thus, the sections that follow begin with a discussion of the importance of the topic why adolescent risky decision making is important and what problems it causes for individuals and society. Then, we turn to the kinds of behavioral change we should hope to achieve to address such problems, by discussing what is rational, adaptive, or good decision making for adolescents. With this ideal of good decision making in mind, we then discuss the main explanatory models of actual adolescent risk taking and the data that bear on major constructs of these models, such as how adolescents perceive risk (e.g., the myth of invulnerability) and what changes in risky decision making occur with development, as adolescents mature and gain experience in the world. In our concluding section, we draw out some key implications of current findings for different approaches to risk reduction and avoidance. SIGNIFICANCE OF THE PROBLEM Why Is Adolescent Risky Decision Making Important? The scientific literature confirms the commonsense belief that adolescence is a period of inordinate risk taking. For example, three million new cases of sexually transmitted infections are identified in adolescents every year in the United States, and more than half of all new cases of HIV infection occur in people younger than 25 (Centers for Disease Control and Prevention, 2004). Moreover, adolescents have one of the most rapidly increasing rates of HIV infection (e.g., 37% more young people were living with AIDS in 2003 than in 1999), with an average of two new young people in the United States infected with HIV every hour (Centers for Disease Control and Prevention, 2004; Rosenberg, Biggar, & Goedert, 1994). Substance use also typically begins in adolescence, as do its adverse health consequences, such as the risky sexual and driving behavior associated with alcohol use (Bachanas et al., 2002; Fergusson & Lynskey, 1996; Strunin & Hingson, 1992; Tapert, Aarons, Sedlar, & Brown, 2001). Motor-vehicle accidents are the leading cause of deaths among those aged 15 to 20 years; 31% of young drivers killed in motor-vehicle crashes in 2003 had been drinking (National Center for Statistics and Analysis, 2003; Turner & McClure, 2003). Although Table 1 indicates decreasing prevalence rates for risky behaviors through 2003, it also shows that progress has recently stalled, that rates remain unacceptably high, and that some behaviors continue to increase (Fig. 1). Apart from the immediate consequences of risky behaviors, adolescents risk behaviors also strongly predict health problems in adulthood (U.S. Department of Health and Human Services, 1994). Behaviors that were begun as voluntary choices INHALANT USERS (%) p< YEARS 8TH GRADE 10TH GRADE 12TH GRADE Fig. 1. Percentage of 8th, 10th, and 12th graders reporting lifetime use of inhalants from 1991 to 2004, showing significant increase among 8th graders (based on Johnston, O Malley, Bachman, & Schulenberg, 2004). to experiment can be perpetuated by addiction (Slovic, 2000, 2001). Although most drinkers do not progress to deep alcoholism, virtually all alcoholics started drinking in adolescence (Bonnie & O Connell, 2004; Chambers, Taylor, & Potenza, 2003; Vaillant, 2003; Vaillant & Hiller-Sturmhöfel, 1996; see Table 2 for data pinpointing adolescence as a period of initial vulnerability to multiple risks). Prevention at the time when use is still a matter of deliberate choice is more successful and less costly and, thus, better for adolescents and for society, than dealing with an established addiction later. Delaying the onset of drinking and reducing the amount consumed decrease the risk of progressing to alcoholism. 2 Delay and reduction also allow the forebrain and other neurological structures that contribute to judgment and behavioral inhibition to mature, which should further reduce unhealthy risk taking (Crone & van der Molen, 2004; Dempster, 1992; Galvan et al., 2006; Hooper, Luciana, Conklin, & Yarger, 2004; Steinberg, 2005; Fig. 2). Similar arguments can be made for postponing sexual activity and certain other risky behaviors. Not only are immediate negative outcomes reduced, but older adolescents bring a more developed brain, as well as greater social and emotional maturity, to risky situations (e.g., Byrnes, 1998; Reyna, 1996). 2 The problems we discuss in this article are not limited to the United States. For example, a recent report debunks the myth that European adolescents have fewer alcohol problems because their cultures teach them to handle alcohol responsibly from an early age; the report shows that a large majority of European countries had higher intoxication rates and binge drinking (five or more drinks in a row) rates among adolescents than the United States. Data collected from 15- and 16-year-olds in 35 European countries showed that European adolescents drink more often, drink more heavily, and get drunk more often than American adolescents do: In the United States, 22% binge drank in the past 30 days; in Denmark, that figure was 60%; in Germany, 57%; in Britain, 54%; in Italy, 34%; and in France, 28%. Intoxication rate in the last 30 days for U.S. adolescents was 18%, compared to 61% in Denmark, 53% in Ireland, 48% in Austria, and 46% in Britain. Only six European countries had lower intoxication rates than the United States. Data from Europe were collected as part of the European School Survey Project on Alcohol and Other Drugs, and the U.S. data were from the Monitoring the Future survey conducted annually among 8th, 10th, and 12th graders in the United States. These data have implications for hypotheses about the effects of accessibility of alcohol and for such public policies as raising drinking ages (see Grube, 2005). Volume 7 Number 1 7

8 Risk and Rationality in Adolescent Decision Making TABLE 2 Adolescent Exposure to Risks and Early Onset of Risk-Taking Behavior Risk/behavior Data Alcohol 40% of adult alcoholics report having initial alcoholism symptoms between the ages of 15 and 19. Car accidents Between the ages of 16 and 20, both sexes are at least twice as likely to be in accidents than are drivers between the ages of 20 and 50. These accidents are the leading cause of adolescent death. Gambling Pathological or problem gambling is found in 10% to 14% of adolescents, and gambling typically begins by age 12. Sexual activity Adolescents are more likely than adults to engage in impulsive sexual behavior, to have multiple partners, and to not use contraception. Younger teens (12 14 years) are more likely to engage in risky sexual behavior than are older teens (16 19 years). STDs Annually, 3 million adolescents contract a sexually transmitted disease. HIV infection is the seventh leading cause of death among 13- to 24-year-olds. Note. Data sources include Bachanas et al. (2002); Chambers & Potenza (2003); Chambers, Taylor, & Potenza (2003); and Turner & McClure (2003). Indeed, the likelihood of engaging in many risky activities is greatly reduced simply by aging. Generally speaking, the prevalence of illegal behaviors is lower during adulthood than during adolescence (Menard, 2002; Moffitt, 1993; National Research Council and Institute of Medicine, 2001). Both arrest data and self-report surveys reveal that crime rates peak during adolescence and young adulthood. The arrest rate for serious violent crimes rises rapidly during the adolescent years, reaches its height at age 18, and drops rapidly thereafter. One reason that violent crime rates fell in the late 1990s was the aging of the population; young people made up a smaller share of the total, producing lower crime rates by default. However, America s juvenile population has grown significantly over the past several years, rising from 13.3 million in 1990 to 14.8 million in 1995 to 15.7 million in The adolescent population is expected to peak in 2007 at 17.3 million. As the gateway to adulthood, adolescence represents a combination of increased accessibility to risk-taking opportunities (e.g., adolescents drive and have less adult supervision) coupled with immature risk attitudes, understanding, and self-regulation (e.g., Byrnes, 1998; Gottfredson & Hirschi, 1990; Reyna, 1996). Crime, smoking, drug use, alcoholism, reckless driving, and many other unhealthy patterns of behavior that play out over a lifetime usually debut in adolescence. Avoiding unhealthy risks or buying time during adolescence before exposure to risks can therefore set a different lifetime pattern. The public-policy implications of these observations are straightforward: Changing unhealthy behaviors in adolescence would have a broad impact on society, reducing the burdens of disease, injury, human suffering, and associated economic costs. What Is Rational, Adaptive, or Good Decision Making for Adolescents? Traditional Behavioral Decision-Making Models Traditional behavioral decision-making models have been widely applied in decision research with both adolescents and adults, and they are the standard against which new behavioral approaches are compared. Following Edwards (1954) original formulation, comprehensive treatment of any decision requires three forms of interrelated research: normative, descriptive, and prescriptive. These characterize, in turn, rational decision making, actual behavior, and interventions that bridge the gap between the normative ideal and the descriptive reality (Bell, Raiffa, & Tversky, 1988; Fischhoff, 2005; von Winterfeldt & Edwards, 1986). (Our use of the term normative may be confusing to social scientists who use the same term to refer to the concept of the average, or norm, rather than the ideal; however, because use of the term to mean ideal is standard in decision research, we have adopted that usage here.) 3 Rationality, in this approach, involves making choices that best realize the decision maker s goals, regardless of what those goals might be (we return to this topic in the next section). Although naturally presented in sequential terms, the three stages are inherently intertwined: Descriptive research about actual behavior can show that normative analyses have mischaracterized decision makers goals (and, thus, seemingly irrational behavior can be seen as rational, given the decision makers goals). Similarly, interventions can test the depth of researchers understanding of descriptive results about actual behavior, showing that supposed causal mechanisms underlying behavior do not respond to valid interventions (and, thus, that researchers need to think harder about the causes of observed behaviors). We discuss the normative analysis of decision making in this section; we then offer a descriptive analysis of what is known about adolescent risky behaviors; finally, we discuss how these analyses lead to prescriptions (including policy implications) for improving adolescent decision making. In the context of adolescent risk decisions, normative analyses ask questions such as what really matters to adolescents when contemplating behaviors, which options give them the best chances of achieving those outcomes, and what information would make those choices clearer? Descriptive research asks 3 What is considered normative or ideal decision making, however, varies from theory to theory, although the classical view has been that the normative ideal is rationality, defined as consistency with the axioms of subjective expected utility theory (discussed in the text) and with the rules of logic and probability. 8 Volume 7 Number 1

9 Valerie F. Reyna and Frank Farley Frontal Lobe Executive Function Planning Reasoning Impulse Control Dorsolateral Ventromedial Thinking ahead and inhibition of impulsive responses Regulation of emotions; learning from experience; weighing risks and rewards Fig. 2. Brain areas and functions showing significant development during adolescence. The frontal lobe (top panel, in pink) continues to mature into the mid-20s; its sub-areas, the dorsolateral and ventromedial areas (bottom panels) are associated with impulsivity, thinking ahead, and other decision factors. questions such as how well do adolescents know what risks they are facing, how well can they anticipate how they will feel if things go wrong, and how well can they control their emotions when they need to think coherently? Prescriptive research asks questions such as do we understand adolescents well enough to help them appreciate the long-term consequences of their actions and do we understand their world well enough to reduce unmanageable social pressures? More formally, the normative analysis of a choice identifies the options in the decision makers best interests, given their goals and the information available to them, all integrated by the application of a rational decision rule. Customarily, that is an expected utility rule, which multiplies the utility (or attractiveness) of each outcome by the probability of its being obtained for each option. In these terms, rationality is a matter of consistency with a set of rules, such as transitivity (e.g., individuals who prefer A to B and B to C should also prefer A to C), because following such rules can be shown to result in reaching the decision makers goals (i.e., maximizing the attractiveness, to that decision maker, of the chosen option; von Neumann & Morgenstern, 1944; Yates, 1990). Whether people actually adhere to such rules or pursue their own best interests is a detail left to descriptive research as opposed to normative analysis. In other words, people might not be capable of engaging in rational decision making, as defined by utility maximization or some other rule, but that should not be confused with what is defined Volume 7 Number 1 9

10 Risk and Rationality in Adolescent Decision Making as normatively ideal (i.e., what prescription should aim to get as close as possible to, even if the normative ideal is never reached). Behavioral decision theory does not tell people what to believe, but it stipulates that rational choices are ones that use decision makers current beliefs in an orderly way. If those beliefs are inaccurate, then the resulting choices need not be optimal. The term subjective expected utility refers to choices based on intuitive probability judgments rather than the best available knowledge. The implication, then, is that decision making cannot be described as irrational when it requires knowledge that the decision maker did not have. Decisions can be wrong because of ignorance, but not necessarily irrational a fundamental distinction when evaluating the decisions of adolescents, who may lack crucial knowledge and experience. Normative analyses also recognize that people may rationally pursue goals that others dislike (e.g., adolescents who care more about good times and social approval than adults think is appropriate). Normative analyses recognize that people may make choices with unhappy outcomes because no better options were feasible (e.g., when dealing with bullying or sexual coercion). Normative analyses recognize that bad outcomes may follow good decisions, when chance intervenes, just as good luck may reward poor choices. Indeed, there is a term in decision analysis, outcome bias, for confusing the quality of decision processes and the consequences of decisions (Ritov & Baron, 1995). Normative analyses of adolescents circumstances can have disquieting results, as when they show adolescents to have the wrong goals or to be trapped in miserable situations with no good choices available to them. However, according to behavioral decision models, such analyses are essential to evaluating adolescents performance and to designing interventions that encourage them to do things that adults want or, to recognizing their contrary aspirations. In this view, good science and good policy require a full analysis of the decisions that people face. A behavioral decision research perspective has little use for assessing people s understanding of facts that are absent from normative analyses, which include critical facts about goals, options, probabilities, and consequences. Descriptive research has, then, a vital role to play in identifying barriers to grasping those critical facts. For example, adolescents may know that safe sex is important but not what that term means (McIntyre & West, 1992). They may know that any unprotected sex runs some risk of pregnancy or sexually transmitted infection, in the sense that you can get it the first time, but may not how to interpret the experience of getting through the first time safely (e.g., are they immune? infertile?; Downs, Bruine de Bruin, Murray, & Fischhoff, 2004). In sum, according to behavioral decision theory, a normative analysis of a choice proceeds by describing it in terms that take full advantage of the best available information on the topic and then combining it with a rational decision process (such as maximizing subjective expected utility), in order to identify the choice that best realizes the decision maker s goals. The behavioral decision perspective traditionally describes rationality in terms of achieving the decision maker s goals, however unacceptable or abhorrent those goals may be to others. Behavioral decision theory incorporates a normative ideal about how decisions should be made but not about what people should want or believe. Regardless of whether all of the tenets of the behavioral decision model are accepted, the tripartite distinction among normative, descriptive, and prescriptive perspectives on decision making is useful: Any program designed to prevent or change risky behaviors should be founded on a clear idea of what is normative (what behaviors, ideally, should the program foster?), descriptive (how are adolescents making decisions in the absence of the program?), and prescriptive (which practices can realistically move adolescent decisions closer to the normative ideal?). Coherence Versus Correspondence Criteria for Rational Decision Making As the reader might surmise, how to characterize a rational decision process has been a matter of debate. There are two sides to this debate, referred to as coherence and correspondence, respectively (Adam & Reyna, 2005; Doherty, 2003). A coherent decision process is internally consistent, often defined with respect to the constraints of relevant formal systems, such as logic or probability theory (e.g., Gilovich et al., 2002). We have already discussed an example of a coherence criterion, namely, that ordering of preferences should obey transitivity. Another example, taken from probability theory, is that people are said to violate coherence when they commit the conjunction fallacy: ranking the conjunctive event A and B as higher in probability than one of its component events (e.g., B ); feminist bank teller cannot be a more likely description of a person than bank teller (Tversky & Kahneman, 1983). Theorists raised questions in the 1990s about whether logical fallacies such as the conjunction fallacy are really errors, speculating that they are instead the product of a mature intelligence that pays attention to the semantics and pragmatics (not just the logical form) of a problem. For example, it was speculated that participants rated a woman as more likely to be a feminist bank teller than a bank teller because they made the pragmatic inference that bank teller must refer to nonfeminist bank tellers. However, research testing these semantic and pragmatic explanations for human performance found them to account for only a small proportion of variance, despite their rhetorical appeal (e.g., Reyna, 1991; Sloman, Over, Slovak, & Stibel, 2003; Stanovich & West, 2000). A few participants interpreted bank teller as referring to nonfeminist bank tellers, but most did not and yet, they still committed the conjunction fallacy. In other words, although people commonly make semantic and pragmatic inferences in everyday life, these purported inferences were generally not related to the logical fallacies participants committed in laboratory tasks illustrating heuristics and biases. 10 Volume 7 Number 1

11 Valerie F. Reyna and Frank Farley Given participants actual interpretation of the materials and tasks, their reasoning was, in fact, flawed. (Empirically supported explanations for this flawed reasoning include an increasing and, mostly adaptive, reliance on gist representations with development.) Thus, when a person s statements or actions seem nonsensical, illogical, or self-contradictory when they lack coherence it raises doubts about his or her rationality. The coherence view of rationality revolves around such questions as the following: Is reasoning logical does it obey the rules of logic? Are decisions internally consistent with one another do preferences exhibit coherence with respect to basic assumptions such as transitivity (i.e., within the same time frame, if A is preferred over B, and B is preferred over C, then A should be preferred over C) or invariance across superficial changes in packaging or phrasing of the same options? Regardless of what adolescents may be capable of, the normative question is whether they should, ideally, be logical or coherent (or whether adults may have to be coherent on their behalf). Although coherence does not guarantee reaching one s goals, it is a minimal condition for doing so. In contrast, the correspondence view of rationality implies that reaching one s goals requires being in touch with reality, making judgments and decisions that produce good outcomes in the real world. It is not too far off to characterize coherence as mainly being about the decision process and correspondence as mainly being about the decision outcome. Correspondence has great appeal in characterizing adolescent rationality because it can be closely linked to desirable goals such as physical and psychological well-being. However, good outcomes are surprisingly difficult to define objectively (whereas logic and probability theory have explicit rules that provide a rationality calculus). Moreover, outcomes do not necessarily reflect the agency of the decision maker. For example, an adolescent or any other decision maker (e.g., a physician) can do everything right and have a bad outcome, and conversely (the outcome bias described earlier). To whit, Winston Churchill (the prime minister) was overweight, drank copiously, and smoked cigars, but he lived to 90, whereas Jim Fixx (the runner) reputedly lived a Spartan life of careful dieting and vigorous exercise, but he was dead at 52. Other factors, such as genetics, can sometimes determine outcomes, despite good or bad decision processes. In the long run and over large samples of people, however, rational decisions ought to lead to better outcomes. Thus, as a matter of public policy, it makes sense to foster rational decision processes (coherence) in adolescence in order to achieve desirable economic, psychological, and public health outcomes (correspondence; Baron & Brown, 1991; Beyth-Marom, Fischhoff, Quadrel, & Furby, 1991; Furby & Beyth-Marom, 1992). Both coherence and correspondence are desirable criteria for rational decision making. As the preceding discussion relating coherence to correspondence suggests, the dichotomy between the view that rationality is fundamentally about formal considerations, such as logical coherence, and the view that it is fundamentally about good outcomes in the real world has been challenged (Doherty, 2003). Recent theories of rationality have integrated these two approaches on the ground that each has limitations and that both coherence and correspondence are therefore necessary for ideal decision making (Reyna & Brainerd, 1994; Reyna, Lloyd, & Brainerd, 2003). Considering adolescent decision making simultaneously from these two perspectives, however, raises a number of foundational questions. First, as we have indicated, traditional behavioral decisionmaking theory holds that rational decision processes help people achieve their goals. From a developmental perspective, we can anticipate that adolescents goals evolve over time. A key question, here, is whether their short-term goals, such as peer acceptance, are incompatible with long-term goals, such as health promotion. Evidence indicates that the answer to this question is yes: Goals vary with age, and short- and long-term goals are often incompatible, even for adults (e.g., Loewenstein & Schkade, 1999). Traditional behavioral decision theory does not offer a means for deciding between present and future goals, although theorists have noted that the pleasures and pains of actual consumption sometimes differ from the anticipation of pleasures and pains when decisions are made (March, 1988). For example, studies indicate that most people rescued from suicide attempts are later glad that they were unsuccessful (e.g., Seiden s, 1978, classic study of survivors who jumped from the Golden Gate Bridge). Figure 3 presents a single, poignant example (for national statistics, see Anderson & Smith, 2003). Owing to their relative lack of experience, adolescents live in a more surprising world than adults do; they are less likely to be able to anticipate their future feelings and goals. If objective evidence indicates that most adolescents who attempt suicide regret the attempt, that most adolescents who refuse chemotherapy ultimately are glad they were forced to take it, and so on, such findings ought to weigh heavily in considerations of what is rational. Note that we emphasize the use of appropriate scientific evidence in judging future goals. Even so, in applying a standard of what most people eventually want, that standard will be wrong for a minority of individuals. (Naturally, all measures should be taken to explain the costs and benefits of options to adolescents so that they have the greatest chance of coming to a mature perspective on their own.) In our view, however, it is worse to ignore good evidence about which course of action will produce better and more satisfying outcomes for the majority of adolescents. Thus we recommend that, in applying correspondence criteria for rationality to adolescents, scientific evidence concerning which behaviors, practices, or policies are likely to produce positive physical and psychological outcomes over the long term must be brought to bear. Again, correspondence criteria are not a substitute for coherence criteria, and our recommendation is that correspondence criteria be modified for adolescents in order to reflect developmental differences in their ability to anticipate future goals and outcomes. Volume 7 Number 1 11

12 Risk and Rationality in Adolescent Decision Making 0.75 PROPORTION OF TIME CAR IN MOTION ALONE WITH FRIENDS ADULTS YOUTHS ADOLESCENTS Fig. 4. Proportion of time car was in motion in a simulated driving task for adolescents (13 16 years old), youths (18 22 years old), and adults (over 24 years old), both alone and with friends in the car, in decisions to run a yellow light (based on Gardner & Steinberg, 2005; Steinberg, 2005). Fig. 3. Letter from a student who attempted suicide, published in the Cornell Daily Sun on September 14, 2005 (Ten years ago, I tried to kill myself in the A lot, 2005). The second foundational question is whether coherence criteria for rationality are appropriately applied to adolescents. That is, are adolescents sufficiently advanced cognitively to be capable of achieving logical coherence? The answer to this question is a highly qualified yes (e.g., Steinberg, 2004; although it should be noted that even adults do not exhibit logical competence reliably across contexts). To observe adolescents underlying reasoning competence, conditions must be optimal, and optimal conditions for making risky decisions are rare in everyday life. For example, using questionnaires and a laboratory behavioral task measuring risk preference and risk taking, Gardner and Steinberg (2005) showed that people made riskier decisions in the presence of peers than they did alone, and this effect was magnified for adolescents and youth compared to adults. The behavioral task mirrored real-life decisions involving reckless driving, such as the decision to run a yellow light and continue through an intersection although, as it was a simulation, no consequential risk was involved (Fig. 4). Note that adolescents were capable of making better decisions and showing more mature risk preferences without peers, evincing what Reyna and Brainerd (1994) termed task variability (the ability to exhibit a higher level of competence under one set of circumstances than under another, superficially different circumstance in which the same competence is relevant). The basic qualitative rules of logic and probability theory are intuitively understood early in development (Falmagne, 1975; Fischbein, 1975; Huber & Huber, 1987; Reyna & Brainerd, 1994), although, again, such competence may not be reliably expressed across contexts at any age. Research suggests that adolescents can be coherent reasoners and can learn from experience about which behaviors lead to good outcomes. Crone, Vendel, and van der Molen (2003), for example, showed that adolescents could learn from positive and negative outcomes in a learning task linked to real-life decision-making competence, although those high in sensation seeking were oversensitive to rewards. In principle, barring temptations with high rewards and individual differences that reduce self-control in other words, under ideal conditions adolescents are capable of rational decision making to achieve their goals (i.e., they can be coherent reasoners who can learn to achieve desirable outcomes). In practice, as we discuss in detail in a subsequent section, there is a considerable gap between ideal reasoning competence and actual performance in the laboratory or the real world. Therefore, much depends on the situation in which the decision is made. In the heat of passion, on the spur of the moment, in unfamiliar situations, when trading off risks and benefits favors bad long-term outcomes, and when behavioral inhibition is required for good outcomes, adolescents are likely to reason more poorly than adults (Amsel, Cottrell, Sullivan, & Bowden, 2005; Crone & van der Molen, 2004; Dempster, 1992; Hooper et al., 2004; Overton, 1990; Overton & Byrnes, 1991; Steinberg, 2004). However, in some situations (e.g., consent for nonemergency medical treatment), conditions that better tap underlying reasoning competence can be arranged (e.g., see Reyna & Brainerd, 1994, for a review). However, as indicated in our earlier discussion of adolescents time horizons, even under good conditions, short-term goals (e.g., not losing hair in chemotherapy) are apt to weigh more heavily than long-term goals for adolescents, relative to adults. As these two questions about adolescent rationality illustrate, healthy decision making is not the same thing as rational or normative decision making as traditionally defined. What is healthy in the narrow sense of promoting psychological and physical well-being may conflict with a decision maker s goals. If a decision maker s goal is to maximize immediate pleasure, for example, many kinds of unhealthy behavior, such as drinking and drug use, could be deemed rational (see Baron, 2003, for an unapologetic defense of the standard view). For example, theorists could make the case that drug use, smoking, or other risky activities maximize utility and are, therefore, rational. Some 12 Volume 7 Number 1

13 Valerie F. Reyna and Frank Farley theorists have attempted to fix this apparent shortcoming in traditional theories (i.e., the prediction that seemingly irrational self-destructive behavior could be construed as rational pursuit of personal goals) by appealing to notions of addiction and temporal discounting (the idea that delayed outcomes are valued less), among other concepts. Most decision theorists still accept the standard view because of the very real difficulties of evaluating other people s personal goals. Again, recent theories have begun to challenge the standard view on this point: Merely assisting people in achieving their own goals no matter how misguided, inconsistent with reality, or self-destructive those goals might be is being questioned by decision theorists and even rejected in some quarters (see Doherty, 2003). This shortcoming of traditional decision theory is another reason why we include correspondence (healthy outcomes) as well as internal coherence as criteria for rationality. In our view, having healthy goals that result in healthy outcomes is an essential aspect of human rationality, and, with respect to policy, society has a justifiable interest in promoting physical and psychological health (Byrnes, 1998; Payne, Bettman, & Johnson, 1993). Despite common usage, healthy decision making is also not the same thing as adaptive decision making in the strict evolutionary sense. According to evolutionary theory, adaptive behavior leads to the preservation and propagation of one s genes. In this sense, smoking is adaptive (or, more precisely, not maladaptive) because most harmful consequences occur after procreation and child rearing have occurred. Similarly, early sexual promiscuity can be considered adaptive from an evolutionary perspective (see Baumeister, 2005, Geary, 2005, and Gigerenzer & Selten, 2001, for more elaborated approaches to selection pressures). The need for a protracted education (and thus deferred procreation) in an industrial society is a recent development historically and would not have been evolutionarily selected or preferred. Therefore, in a modern industrial economy, early pregnancy and child rearing is maladaptive but only in a metaphorical sense because it does not promote the physical or psychological health of oneself or one s offspring. Evolutionary theory has been widely touted as a guide to what should be considered smart behavior in decision making (e.g., Gigerenzer, Todd, & the ABC Group, 1999). However, evolutionary theory about what is adaptive offers little guidance for public policy in the arena of adolescent decision making. On the contrary, some behaviors that are adaptive from an evolutionary perspective, such as early procreation, are precisely the behaviors that are not healthy in a modern society. Thus, adolescent decision making represents an important counterexample to assertions of contemporary evolutionary theories that what is adaptive in the evolutionary sense is also rational, healthy, or desirable behavior from either a theoretical or a public-policy perspective. Self-destructive behavior could be construed as adaptive from an evolutionary perspective and perhaps be rationalized as the pursuit of personal goals and, nevertheless, be unhealthy behavior that policy ought to discourage. Given expected conflicts in adolescence between personal goals and healthy goals, the question arises as to whether it is ever rational to engage in unhealthy behaviors. The assumption that is implicit in this question is that people in their right minds or who are thinking logically would not intentionally harm themselves. The premise that one prefers not to be hurt and yet is doing something to hurt oneself does not make logical sense; the preference and the action seem to contradict one another. Assuming that people are in their right minds and thinking logically, the implication is that people who engage in unhealthy behaviors must not realize that the behaviors are harmful (or fully realize the extent of harm), or that there are other benefits to the behaviors that may not be readily apparent. Thus, there are ways behavior that seems nonsensical can, after the fact, be made sense of e.g., by inferring compensating rewards or benefits. An argument for welfare reform, for example, was that the existing system encouraged adolescents to become pregnant because welfare benefits provided them with income (despite a lack of education) and an independent domicile away from their parents. Others have speculated that adolescents have babies so they can have someone to love or someone who will look up to them, especially if there are few other possible sources of love or admiration. There is no conclusive evidence to support either of these speculations. Nevertheless, it is important to acknowledge the benefits that adolescents derive from engaging in risky behaviors, such as gaining social acceptance from peers (and there is evidence that perceived benefits drive risk taking in adolescence; see below). The main problem with such accounts is that it is always possible, post hoc, to conjure up benefits that make any behavior appear rational. We would argue, instead, that it is possible to acknowledge the benefits that adolescents derive from specific behaviors and, thus, explain their motivations without necessarily characterizing their decisions as healthy, adaptive (in the context of modern society), or rational overall. (Again, we draw a distinction between evolutionarily adaptive based on past selection pressures, which encouraged specific kinds of risk taking with adaptive in the context of an individual s mental and physical health in modern society.) Many of the behaviors we have discussed smoking, drug use, and unsafe sexual activity appear to offer immediate pleasures, whereas any adverse outcomes are generally longer term (e.g., Herrnstein, & Prelec, 1992). Another prominent example comes from the recent epidemic of adolescent obesity: Eating tasty fast foods (immediate pleasure) runs counter to the long-term goal of physical health (e.g., avoiding well-known delayed outcomes such as diabetes and cardiovascular disease). Temporal discounting, the tendency to weight immediate outcomes more heavily than future outcomes, is a robust characteristic of adult preferences in decision making (Loewenstein & Elster, 1992). The pleasure of receiving $100 now is greater than that of receiving the same $100 in a month or a year. In order to account for real behavior, models of experiential learning about risks also weight recent outcomes more heavily than those more Volume 7 Number 1 13

14 Risk and Rationality in Adolescent Decision Making removed in time (i.e., outcomes more removed in the past; Busemeyer & Stout, 2002; Busemeyer, Stout, & Finn, in press; Hertwig, Barron, Weber, & Erev, 2004; Rottenstreich & Hsee, 2001; Weber, Shafir, & Blais, 2004). Because the environment changes, it makes sense to weight recent information more heavily than old information. Thus, the argument could be made that a generally adaptive tendency to weight recent more than distant outcomes occasionally backfires by encouraging immediate feel-good behaviors, such as smoking, drug use, overeating, and risky sexual behavior. However, there is no evidence that differential weighting of immediate outcomes is adaptive in the sense that it promotes healthy outcomes in a modern society (but cf. Ainslie, 2001). On the contrary, planning ability, future orientation, lack of impulsivity, and delay of gratification have each been linked to socially desirable outcomes, such as higher educational attainment and lower propensity for risky and antisocial behaviors (Baron & Brown, 1991; Grisso et al., 2003; Metcalfe & Mischel, 1999; Fig. 5). Moreover, animals seem to have a shorter time horizon and are more impulsive than are human children, who, in turn, are more impulsive than human adults (Ainslie, 2001; Metcalfe & Mischel, 1999; Rachlin, 2000). Hence, both the ontogenetic and phylogenetic evidence favors the long view as more advanced than the short-term perspective, and as more likely to lead to positive outcomes in a modern society. Although arguments about rationality have historically been axiomatic (e.g., von Neumann & Morgenstern, 1944; Savage, 1972) or philosophical (e.g., Harman, 1986), some researchers have argued that developmental data should play a central role (Jacobs & Klaczynski, 2002, 2005; Reyna & Brainerd, 1994, 1995; Reyna, Lloyd, & Brainerd, 2003). According to the latter argument, changes in behaviors with age and experience (developmental progress) can, like data about successful outcomes, be used as objective input into judgments of rationality. This recent use of developmental data should be distinguished from the use of anecdotes or suppositions about the nature of development to buttress philosophical claims about rationality, a practice that has a long tradition (e.g., Quine, 1964). For instance, laboratory research has shown that children are risk takers and risk taking generally decreases, especially for higher levels of risk, in the period from childhood to adolescence to adulthood (Levin & Hart, 2003; Reyna, 1996; Reyna & Ellis, 1994; Reyna & Mattson, 1994; Rice, 1995; but see Schlottmann & Tring, 2005). These developmental studies included risks that involved both gains and losses, and children were more likely to take risks overall (i.e., the results were not limited to taking risks involving gains). Although this pattern conforms to longstanding conventional wisdom about risk-taking propensity among youth, the demonstration of decreased risk taking with age under controlled laboratory conditions seems to contrast with the view that risk taking increases in adolescence relative to childhood (e.g., Dahl & Spear, 2004; Spear, 2000). The contrast is more apparent than real, however. The laboratory pattern (although FUTURE ORIENTATION to to to to AGE Fig. 5. Future orientation at different ages. Participants rated responses to the item, I would rather save my money for a rainy day than spend it now on something fun (based on Grisso et al., 2003). qualified by individual differences) has been replicated and, ironically, suggests that adolescents preference for risks declines during the period in which exploration and opportunity (and thus, risk-taking behaviors) increase. The developmental trend in risk taking in laboratory tasks is not subject to obvious alternative explanations, such as effects of social-motivational factors, rather than changes in risk preferences per se. For example, adolescents might not prefer to take risks in the real world but might do so anyway to impress their friends. The aforementioned experiments did not involve the presence of peers or other similar social factors that are known to be influential but that would cloud interpretation of the results regarding risk preferences. Children and adolescents prefer to take risks more than adults do, even when peers are not present to egg them on. This developmental trend cannot be directly compared to actual rates of risk taking in the real world, which are confounded by such factors as opportunity and thus do not necessarily reflect underlying risk preferences (but see Moffitt, 1993, and Spear, 2000, who speak to the important issue of the confluence of risk-taking propensity with real-world opportunity). In the laboratory, one can control effectively for opportunity but less effectively for affect, emotional valence, social factors, and so on. Ethical constraints, for example, provide an upper limit on manipulations of negative affect. The absence of serious costs or consequences in laboratory risk-taking tasks could distort the estimate of underlying risk preferences based on these tasks. This issue applies generally to research using laboratory tasks that involve no serious consequences for performance and no direct representations of real-world health decisions, a classic hot-house phenomenon (i.e., an artificial environment that might not extrapolate to the real world; Agnew & Pyke, 1994; exceptions include work by Slovic, Peters, and colleagues, e.g., Finucane, Peters, & Slovic, 2003). One cannot assume that laboratory behavior does not generalize to the real world, and there are numerous examples of successful transfer (e.g., see Brainerd & Reyna, 2005; Parker & Fischhoff, 2005; Yechiam, Busemeyer, Stout, & Bechara, 2005), but neither can 14 Volume 7 Number 1

15 Valerie F. Reyna and Frank Farley the relation be assumed to be necessary. Nevertheless, developmental trends in laboratory risk taking can inform theorists about components of the psychology of risky decision making that then combine with other affective, social, and motivational factors in the real world. Therefore, all other factors being equal, the developmental trend of decreased risk taking with age (mirrored in the realworld data) suggests that greater risk aversion is adaptive (in the broad sense) or rational, and that decision processes that support this aversion are more advanced than those that support risk taking. This developmental shift in greater risk aversion with age does not mean that risk taking is never rational, however. In particular, in situations of deficit (loss), deprivation, starvation, or when one has nothing to lose, risk taking may offer the only means of improving one s situation, as research on foraging in animals indicates (Weber et al., 2004). Indeed, certain types of risk taking can have highly positive features and consequences that will be discussed later (Farley, 2001). However, research has shown that across situations of gain and loss, the global tendency to avoid risk increases from childhood to adulthood, and this robust trend cannot be ignored in deciding which behaviors and decision processes are likely to be rational. (We should remind the reader that we endorse both coherence and correspondence, or promoting healthy outcomes, and that these are separate and equally important considerations in judging rationality.) In short, we argue that empirical evidence about physical and mental health outcomes of behaviors, as well as developmental trends in behaviors, are relevant to claims about adolescent rationality. We cannot leave the topic of rationality without mentioning the most recent theoretical development. Current theories of rationality emphasize dual processes in reasoning and decision making, with two corresponding systems of rationality (Chaiken & Trope, 1999; Epstein, 1994; Kahneman, 2003; Klaczynski, 2005; Sloman, 2002; Stanovich & West, 2000; similar precursor dual theories include Piaget s and Freud s). The first system is fast, associative, and intuitive, whereas the second one is slow, deliberative, and analytical. Theorists speculate that the intuitive system is older evolutionarily and allows the organism to respond quickly to predators but produces reasoning biases and fallacies, whereas the analytical system avoids such biases and fallacies but incurs the burden of lengthy deliberation. Another model, fuzzy-trace theory, draws on evidence for independent gist and verbatim-memory representations of experience, but differs from other dual-process models in emphasizing that there are degrees of rationality and that intuition is an advanced form of reasoning; such claims are based on empirical evidence comparing decision making by children and adolescents to that of adults and decision making of adult novices to that of experts (Reyna, 2004a, 2005; Reyna & Brainerd, 1991b, 1995; Reyna et al., 2003; Reyna, Lloyd, & Whalen, 2001). (As we discuss in greater detail in the next section, intuition fuzzy, gist-based thinking that unfolds in parallel to analytical deliberation and in which few dimensions of information are processed when making a decision can be argued to be an advanced form of reasoning for theoretical and empirical reasons, including the observation that such thinking appears to increase with age and expertise.) 4 Dual-process approaches provide an explanation for why human decision making seems simultaneously impulsive and reflective, intuitive and analytical, qualitative and quantitative. The essential requirement for any scientific theory, however, is that it is predictive rather than merely superficially descriptive (e.g., a list of dichotomies), rhetorically compelling, or even explanatory (cf. Gigerenzer & Regier, 1996). Making novel and counterintuitive predictions is the highest and most diagnostic test of any theory that purports to apply to reality. As we discuss in the next section, dual-process models differ in their emphasis on underlying mechanisms and prediction, as opposed to descriptions, of behavior. In summary, our consideration of rationality leads to the following conclusions: Long-term physical and psychological outcomes, or correspondence criteria, matter in judging behavior Adolescents are capable of coherent decision processes under circumscribed conditions, and rational (coherent) decision processes offer hope of good outcomes in the aggregate and over the long haul Evolutionary arguments must be made with caution and do not necessarily support the achievement of physical and mental health in an industrialized society Personal goals can be at odds with achieving physical and mental health pointedly, for adolescents engaging in risky behaviors Unhealthy behaviors cannot be rationalized after the fact by inferring compensating benefits (although perceived benefits motivate adolescent risk taking, as we show later) Recent rewards are overweighted, especially by adolescents, but planfulness and delay of gratification generally produce better outcomes Developmental trends in behavior can join outcomes data in providing independent evidence of rationality, on the premise that behaviors that increase with age and experience are generally more advanced 4 By intuition we mean fuzzy, impressionistic thinking using vague gist representations, but we distinguish mindless impulsive reaction from insightful intuition that reflects understanding of a situation or decision. Thus, there are two kinds of fast and simple ways of thinking: a stupid kind that represents the most primitive form of thinking and a smart kind that represents the highest form of thinking, insightful intuition. In the foundations of mathematics, intuition is a similarly advanced form of thinking (Reyna & Brainerd, 1991a). Using gist as a core concept, fuzzy-trace theory emphasizes meaning, implying that successful interventions to reduce risk that instill insightful intuition must focus on understanding rather than merely on persuasion or memorization of verbatim facts (see Reyna et al., 2005). Work conducted under the aegis of gestalt theory, a formative influence on fuzzy-trace theory, showed that, in contrast with rote learning, understanding promotes transfer of learning; this suggests that health curricula should promote deeper understanding in order to improve transfer of learning from the classroom to real-world decision making. Volume 7 Number 1 15

16 Risk and Rationality in Adolescent Decision Making Dual-process theories are the most recent approach to encompassing the high and low levels of rationality that characterize human behaviors and, often, the same individual; but predictive theories are required in order to develop effective strategies for prevention and intervention that reduce unhealthy risk taking EXPLANATORY MODELS OF ADOLESCENT RISK TAKING Reasoned, Reactive, and Intuitive Decision Making Most models of adolescent risk taking assume the traditional kind of rational decision process that we have discussed: one that is goal oriented (i.e., directed at reaching personal goals) and logically coherent. According to the behavioral decisionmaking perspective, for example, options are considered, consequences are evaluated, and a decision is made. People are assumed to evaluate options by assessing probabilities, weighting values, and integrating them in order to make a choice all quintessentially cognitive activities. An expanded version of this perspective adds emotional, social, and developmental factors to explain decision making (Fischhoff, 2005). If decision makers care about how other people evaluate their choices, for example, that consideration then becomes another factor in the calculation of costs and benefits. Other rational models include the health-belief model, protection-motivation theory, the theory of reasoned action, the theory of planned behavior, and problem-solving approaches (e.g., Greenberg, Kusche, Cook, & Quamma, 1995; Shure, 2003). Each of these models incorporates mechanisms to explain how people actually make decisions (a descriptive focus) and, to varying degrees, implications of these mechanisms for improving decision making (a prescriptive focus). The aim of the problem-solving approaches, for example, is to develop emotional and social competence, and they encompass such skills as means-end thinking, resistance to peer pressure, seeking help, and generating alternative solutions to problems. Although not all of these problem-solving models have been evaluated with respect to outcomes in adolescence, they constitute instructive attempts to convert reactive and impulsive decision makers into rational, deliberative, and socially competent ones (see also Furby & Beyth-Marom, 1992). A review of problem-solving approaches is beyond the scope of the present article, but evidence of effects in reducing aggressive behavior, as well as other risk-reduction outcomes, has been obtained (see, for example, Greenberg, Domitrovich, & Bumbarger, 2000; Romer, 2003; U.S. Department of Health and Human Services, 2001). As noted earlier, results demonstrating effectiveness suggest that assumed explanatory mechanisms have merit. The health-belief model can be roughly understood as an instantiation of a behavioral decision-making perspective in a health context (Becker, 1990; Byrnes, 1998). The model s components are used to explain why people engage in healthpromoting (or destructive) behavior and, thus, has implications for interventions. The model s components are (a) a person s assumed goal of achieving health (e.g., avoiding or curing illness), (b) perceived vulnerability to health threats, (c) perceived severity of health threats, (d) beliefs that specific behaviors will promote health or cure illness (e.g., beliefs about benefits and barriers to engaging in behaviors to achieve health) and (e) environmental cues to the actions or behaviors that are believed to be effective in achieving health. Broadly construed, the purview of the model includes smoking, dieting and eating disorders, drug and alcohol consumption, and other healthrelated risky behaviors. For example, according to this model, adolescents would be expected to stop smoking if they perceive that the health threats posed by smoking are great (e.g., inability to compete athletically if they continue smoking), that those threats apply to them (e.g., they have asthma and so will be more likely to experience shortness of breath), that the benefits to quitting are significant (e.g., they are on the track team), and that the barriers to quitting (e.g., addiction) are surmountable. Protection-motivation theory is a variant of the health-belief model (e.g., Rogers, 1983). Protection motivation refers to the motivation to protect oneself against a health threat and is usually measured as the intention to adopt some recommended action. Its constructs include perceived vulnerability and severity, response efficacy (the belief that the recommended action is effective in reducing the threat), and perceived self-efficacy (the belief that one can successfully perform the recommended action). That is, people will have a stronger intention to adopt the recommended action to the extent that they believe the threat is likely, that the consequences will be serious if the threat occurs, that the recommended action is effective in reducing the severity of the threat, and that they are able to carry out the recommended action (Sutton, 2001). Two meta-analyses of protection-motivation-theory studies have been conducted and both supported the constructs as predictors of intentions or behaviors (Floyd, Prentice-Dunn, & Rogers, 2000; Milne, Sheeran, & Orbell, 2000). Factors such as perceived vulnerability to and severity of health threats, as well as perceived barriers and benefits to engaging in health behaviors, have been found to be correlated with health behaviors (for a review, see Janz & Becker, 1984). Thus, there is empirical support for the health-belief model (and its variants, including protection-motivation theory). However, many other factors affect health behaviors (i.e., the variance accounted for by the health-belief model is fairly low), and such factors as perceived vulnerability and severity are only weakly correlated with health behaviors (Byrnes, 1998). Although health models are criticized for being narrow, per our discussion of rationality and adaptive decision making, the models can be extended to other behaviors that are only metaphorically adaptive in the sense that they promote physical and mental well-being (e.g., applying to colleges, which involves risk). 16 Volume 7 Number 1

17 Valerie F. Reyna and Frank Farley Moreover, all human behavior is multiply determined, and it is no mean feat to obtain statistically significant effects that predict important health behaviors (although prediction is often limited to statistical association rather than active manipulation of factors). Also, the health-belief model provides obvious entry points for attitude change; if vulnerability is perceived as low (but is objectively high), adolescents should be taught how vulnerable they are and so on with each of the factors. Because decision making is assumed to be conscious and deliberative, explicit instruction ought to make a difference, according to this model, and knowledge ought to be related to behavior (e.g., knowledge of HIV/AIDS risk factors has been found to be negatively associated with adolescent sexual risk-taking, although null effects have also been reported, and knowledge is often insufficient to change behavior; Crisp & Barber, 1995; Kotchick, Shaffer, Forehand, & Miller, 2001; cf. Dudley, O Sullivan, & Moreau, 2002). In our zeal to acknowledge unconscious or nondeliberative effects on behavior, we should not ignore the fact that explicit instruction about vulnerability, severity, benefits, and barriers is sometimes effective in changing behavior. 5 The main difficulties with these models are, first, that they are primarily supported by correlational evidence; they do not really predict outcomes in the sense that underlying mechanisms are understood and have been actively manipulated in experimentation to establish cause effect relations (Kershaw, Niccolai, Ethier, Lewis, & Ickovics, 2003). (Protection-motivation theory has been tested extensively using experimental designs, but these experiments generally do not explore the mechanisms that underlie the constructs.) By correlational, we mean any study that does not involve experimental manipulation of factors, including studies using complex multivariate analyses and statistical controls. Unfortunately, having large sample sizes with many variables that are correlated with one another does not compensate for the absence of a predictive process model of risky decision making. Statistical controls or quasi-experiments are not sufficient to demonstrate causality (Reyna, 2004b). From a practical perspective, this means that, without experiments that support conclusions about causation, programs predicated on correlational studies may nevertheless be ineffective. 5 According to traditional deliberative models of risky decision making, explicit instruction about vulnerability, severity, benefits, and barriers should be effective in changing behavior. One might question, however, whether unconscious antecedents of behavior can be influenced by interventions, which would seem to require conscious reflection. However, this assumption highlights a core difference between deliberative (or computational) and fuzzy-trace models of reasoning and decision making. In the latter model, advanced gist-based reasoning and decision making is often (although not necessarily) unconscious. Indeed, according to that model, the aim of interventions should be to make such thinking unconscious and automatic through practice at intuitively grasping the bottom-line gist (or meaning) of risky situations, and then rapidly retrieving and implementing risk-avoidant values from long-term memory, without conscious deliberation about pros and cons (e.g., Adam & Reyna, 2005; Reyna, Adam, Poirier, LeCroy, & Brainerd, 2005). Second, health-belief models do not account for the unconscious or irrational decision making that seems to be the source of much trouble in adolescence (i.e., impulsive or reactive decision making). As we discuss in connection with the entire class of models that assume rationality as deliberative and analytical, it seems doubtful (relevant data are presented below) that most factors that affect risky decision making are ones that adolescents are consciously aware of (and can report) and that adolescents combine those factors logically and objectively. In other words, it is questionable whether problem behavior in adolescence is exclusively the result of a rational cost benefit analysis (but see Reyna, Adam, Poirier, LeCroy, & Brainerd, 2005). The rational agent hypothesis is a prominent feature of Fishbein and Ajzen s (1975) theory of reasoned action linking beliefs, attitudes, norms, intentions, and behaviors which was later followed by the theory of planned behavior (e.g., Ajzen, 1991; Ajzen & Fishbein, 1980). In both theories, behavioral intention is the immediate antecedent to action (Gibbons, Gerrard, Blanton, & Russell, 1998). Attitudes are the overall affective and instrumental evaluations of performing the behavior. Subjective norms refer to social pressures to perform or not to perform a behavior (e.g., beliefs that parents disapprove or that peers approve of a behavior such as adolescents having sex). The main added construct in the theory of planned behavior is the idea of perceived behavioral control, conceived as a combination of self-efficacy (confidence or sense of ease in performing a task) and controllability (i.e., a sense that the behavior is up to me ; see Rhodes & Courneya, 2004). Perceived behavioral control encompasses perceived resources, skills, and opportunities (Ajzen, 1991). Because behaviors are assumed to be intentional, they involve some degree of premeditation or planning. Behaviors that are not completely volitional are predicted by incorporating perceptions of control as an additional predictor of intention (Ajzen, 1991). These theories have been supported empirically, having effectively predicted health-promoting behaviors such as condom use (Fisher, Fisher, & Rye, 1995) and health screening (McCaul, Sandgren, O Neill, & Hinsz, 1993; see Conner & Sparks, 1996; Sheppard, Hartwick, & Warshaw, 1988, for reviews). A metaanalysis of the theory of reasoned behavior indicated that behavioral intentions accounted for 38% of the variance on average in studies of health behavior (van den Putte, 1993). A meta-analysis of the theory of planned behavior produced a similar estimate of 31% (Armitage & Conner, 2001). As Gibbons et al. (1998) pointed out, however, Not all behaviors are logical or rational... It would be hard to argue that behaviors that impair one s health or well being, such as having sex without contraception when pregnancy is not desired or drunk driving, are either goal-directed or rational.... Nonetheless, these behaviors are common, especially among young persons (p. 1164). Thus, as might be expected, health-impairing behaviors such as substance use, drunk driving, and smoking, as opposed to behavioral intentions, are sometimes not as well predicted by these theories (Morojele & Stephenson, 1994; Stacy, Bentler, & Volume 7 Number 1 17

18 Risk and Rationality in Adolescent Decision Making Flay, 1994). Nevertheless, statistically significant associations between intended and actual frequency of substance use among adolescents have been obtained, supporting some degree of intentionality in these behaviors (Ajzen, 1989; Downey & O Rourke, 1976; Huba, Wingard, & Bentler, 1979; Swisher & Hu, 1983; Wolford & Swisher, 1986). Taken as a whole, these results suggest that risk-taking behaviors in adolescence can originate either intentionally or unintentionally, with each type of risk taking calling for a different kind of intervention. For example, intentional risk taking might be better modified by explicitly addressing such factors as perceived risks, benefits, and norms (e.g., that fewer peers are sexually active than believed). Unintentional risk taking, however, has been described as reactive, or as behavioral willingness, in contrast to behavioral expectations or intentions (Gibbons et al., 1998; Gibbons, Gerrard, & Lane, 2003). Behavioral expectation (the perceived likelihood of engaging in a behavior) is a modified measure of behavioral intention (planning to engage in a behavior) that is more inclusive and, thus, captures more variance in behavior. Behavioral willingness is an even more sensitive measure of susceptibility to risk taking and one that explains unique variance because adolescents are willing to do riskier things than they either intend or expect to do (Gibbons et al., 2003). Thus, there is a group of adolescents who indicate that they would be willing to engage in specific risk-taking behaviors but deny that they are expecting or intending to engage in those behaviors, and studies show that they go on to engage in those behaviors more often than do those who deny willingness. Adolescents who fall into the latter group (willing but without conscious intentions) are likely to be especially at risk because they do not take precautionary measures (e.g., carry condoms or arrange for a designated driver). Unintentional risk taking would be expected to be reduced by such measures as adult supervision or monitoring, because these remove opportunities to react to temptations. Indeed, amount of unsupervised time has been found to predict adolescent risk taking in a variety of domains (e.g., Chassin, Pillow, Curran, Molina, & Barrera, 1993; Crosby et al., 2001; Lahey, Gordon, Loeber, Stouthamer-Loeber, & Farrington, 1999; Lynch, Coles, Corley, & Falek, 2003; Millstein & Igra, 1995; Vitaro, Brendgen, Ladouceur, & Tremblay, 2001). Approaches such as supervision address the criticism that helping young people avoid traps such as trying smoking (and becoming addicted) requires more than inculcating rational decision skills. It is not sufficient to encourage sound thinking and problem solving; the environments in which adolescents develop must also be modified to remove opportunities for unhealthy risk taking when adolescents are not ready to handle them. Note that supervision protects young people from experiencing negative feedback (because they are not put in a position to take risks and, thus, experience bad consequences). Because of this lack of opportunity to learn self-regulation and other selfcontrol strategies, some theorists (e.g., Byrnes, 1998) have suggested that sheltered, inexperienced (p. 153) children would be at higher risk (Byrnes self-regulation model). This hypothesis would be supported by a curvilinear relationship between amount of monitoring and unhealthy risk taking very low and very high monitoring both producing greater unhealthy risk taking but moderate amounts of monitoring producing adaptive behavior. However, within the ranges of amount of monitoring that have been studied, there is little evidence for such a relationship (cf. Shedler & Block, 1990). The reason for the effectiveness of supervision seems clear: Adolescents who either intend or are willing to engage in unhealthy risk taking are thwarted by thorough monitoring. The prototype/willingness model, which incorporates the behavioral-willingness construct, has been supported by studies showing that much adolescent risk behavior is not planned and that willingness and intention are related but independent constructs that separately predict risk behavior (Gibbons et al., 1998; Gibbons et al., 2004). The prototype/willingness model can be thought of as an extension of the theory of reasoned action, retaining such elements as social norms and behavioral intentions but broadening the theoretical purview to unintentional behavior by using new constructs such as willingness. The model suggests that intentions and expectations become better predictors of behavior as maturity increases, whereas with maturity the predictive power of willingness decreases. In the prototype/willingness model, prototypes are images of risk takers and non-risk takers, as well as images of self, that have been found to motivate behavior. Interestingly, overall favorability of images (e.g., of substance users) predicts risky behavior better than do specific attributes described by subsets of adjectives. (For example, when asked to describe a typical teenage smoker, only the overall positivity or negativity of the described image matters; the details do not predict risk-taking behavior.) These data are also predicted by fuzzy-trace theory, which holds that risky behavior is governed by fuzzy gist representations of categories of people, objects, and events (or by values and principles that are cued in context) rather than by verbatim details (Reyna 2004a; Reyna & Adam, 2003; Reyna et al., 2005). As Gibbons, Gerrard, and Lane (2003) write, it is not specific characteristics of the images that motivate behavior (as goals), but rather the general impression of the type of person who engages that is influential (p. 127). Thus, dual-process models, such as the prototype/willingness model and fuzzy-trace theory, identify two divergent paths to risk taking: a reasoned route and a reactive route. Although fuzzy-trace theory shares characteristics of the prototype/willingness model, it differs from that model in important ways. Both models explain risk taking that is not reasoned in the usual sense of that term, as well as explaining the declining tendency with age to react without thinking (Steinberg, 2003; for discussions of the role of inhibition in fuzzy-trace theory, see Reyna, 1991, 1995; Reyna & Mills, in press). However, fuzzy-trace theory assigns a central role in advanced 18 Volume 7 Number 1

19 Valerie F. Reyna and Frank Farley decision making to intuition in contrast to the prototype/ willingness model as well as to traditional developmental and decision theories, in which advanced decision making is precise, analytical, and deliberative. The core assumptions of fuzzy-trace theory are based on research in memory, judgment, and decision making, taking into account social, cognitive, affective, and developmental factors (for overviews, see e.g., Reyna & Brainerd, 1995; Reyna et al., 2003). According to the theory, people encode multiple mental representations of their experience, ranging from precise verbatim representations that incorporate surface detail (e.g., exact wording of a risk communication, such as a product label; Reyna & Adam, 2003) to fuzzy gist representations that incorporate the essential meaning of an experience, gleaned through the filter of affect, culture, worldview, education, developmental level, and other factors known to affect semantic interpretation and inference (e.g., Reyna & Kiernan, 1994, 1995). Evidence from experiments with children, adolescents, and adults has shown that such verbatim and gist representations are encoded, stored, and retrieved independently (see Reyna, 2005; Reyna & Brainerd, 1995, for reviews). Decision makers recognize the gist of a risky situation (often multiple gists of that situation) based on prior experience, and simultaneously encode its verbatim representation. Verbatim representations rapidly fade, and judgment and decision making are instead governed by a fuzzy processing preference (i.e., decision making preferentially operates on the gist representations, not on the verbatim ones). This tendency to base decisions on simple qualitative gist increases with age, experience, and expertise, as demonstrated by research with children and adults. As decision making becomes cognitively simpler (but not simpleminded) and gist-based, the tendency to take risks for example, in tasks involving choosing between sure things and gambles generally declines (Levin & Hart, 2003; Reyna, 1996; Reyna & Mattson, 1994; Rice, 1995). Figures 6 and 7 display developmental differences in risk taking, especially for higher levels of risk, for decisions involving both gains and losses (Levin & Hart, 2003, extended the research to adults and showed a child-to-adult decline in risk taking). Experimental evidence indicates that young children roughly multiply probabilities by magnitudes of outcomes (e.g., the number of prizes associated with each possible outcome) in decision tasks, quantitatively combining two dimensions (e.g., Schlottmann, 2000, 2001; Schlottmann & Anderson, 1994). On the same tasks, this quantitative focus slips to one dimension (outcomes) as children get older; adult performance has been shown to not be based on the quantities at all but rather on their qualitative gist (e.g., winning some prizes versus maybe winning some prizes or maybe winning none; Reyna & Brainerd, 1991b, 1994, 1995). (These conclusions are based on actively manipulating factors in experimental tests, presenting many decisions per child and using ratings and other preference measures, as opposed to being based on behavior on a few choice trials.) Thus, it % RISK TAKING RISK PREFERENCE (A) (B) RISK LEVEL PRESCHOOL 2ND 5TH RISK LEVEL PRESCHOOL 2ND 5TH Fig. 6. Choice proportions for the risky, as opposed to sure, option (panel A) and risk-preference ratings on a 7-point happy-face scale (i.e., children pointed to smiley faces that varied from neutral to very happy; panel B) for 3 risk levels across gain and loss decisions for 44 preschoolers, 33 second graders, and 47 fifth graders; ratings for sure choices were multiplied by 1 so that ratings could vary from strongest preference for the gamble (17) to strongest preference for the sure option ( 7; based on Reyna & Mattson, 1994). is young children who demonstrate sophisticated quantitative competence, trading off the magnitude of rewards against the magnitude of risks, modulating their preference for risk according to the overall quantitative value of the options (obtaining these findings requires highly sensitive methodologies, but the results have now been replicated in several laboratories). Adults, in contrast, engage in simpler (but not simplistic) decision processes (see also Table 3). The empirical evidence from laboratory studies supports the conclusion that gist-based intuition produces risk avoidance, but deliberation weighing of alternatives encourages risk taking, and gist-based intuition is associated with maturity (e.g., Reyna et al., 2005). Representations alone do not determine decision making; retrieval of values and their implementation in context are also Volume 7 Number 1 19

20 Risk and Rationality in Adolescent Decision Making RISK PREFERENCE % RISK TAKING (A) (B) RISK LEVEL PRESCHOOL 2ND 5TH RISK LEVEL PRESCHOOL 2ND 5TH option. Variability in situational cues, in part, explains task variability and apparent instability of preferences and decisions. Compared to adults, adolescents have less experience with situational cues concerning risk, and thus they are less likely to recognize danger or to immediately think of consequences. Fuzzy-trace theory, therefore, emphasizes reactions to cues in the environment, although the mental processes of advanced decision makers have been distinguished from merely acting on impulse (e.g., Reyna, 1991, 1995). Advanced decision makers rapidly home in on the essential gist, ignoring verbatim detail and irrelevant cues. For example, studies of physicians making risky decisions in emergency rooms have demonstrated that, when they make decisions in their domain of expertise, more knowledgeable individuals (e.g., cardiologists) process fewer dimensions of information and do so more qualitatively (consistent with using gist representations) than do those with less knowledge and training (and yet, more knowledgeable physicians medical decisions are more accurate; Reyna & Lloyd, in press; Reyna et al., 2003; see also Dijksterhuis, Bos, Nordgren, & van Baaren, 2006). Evidence from these and other studies suggests that more advanced decision makers (adults compared to children or experts compared to novices) automatically encode the gists of risky situations, retrieve risk-avoidant values that are appropriate to those situations, and smoothly apply those generic values to the specific situations. The difference between advanced decision makers and impulsive reactors lies in the ability of the former to quickly react to a small number of relevant cues, as opposed to reacting to misleading or irrelevant lures and other sources of temptation. Fig. 7. Choice proportions for the risky, as opposed to sure, option (panel A) and risk-preference ratings on a 7-point happy-face scale (i.e., children pointed to smiley faces that varied from neutral to very happy; panel B) for 3 risk levels across gain and loss decisions for 44 preschoolers, 51 second graders, and 51 fifth graders; ratings for sure choices were multiplied by 1 so that ratings could vary from strongest preference for the gamble (17) to strongest preference for the sure option ( 7; based on Rice, 1995; see also Reyna, 1996). critical. As a result of acculturation, children acquire values that they endorse and store in a vague form in long-term memory (e.g., life is better than death; it is better to have a relationship than to be alone). Depending on the cues in the situation, people retrieve their values from long-term memory and apply them to the gist representation of the situation (fuzzy-trace theory has a detailed retrieval model, which has been formalized using mathematical models whose parameters have been tested individually and collectively for goodness of fit to actual data; e.g., Brainerd, Reyna, & Mojardin, 1999). Affect is one important contextual cue, among others, that prompts retrieval of values. In the example of a choice of a sure thing or a gamble with varying prizes, people generally retrieve such values as more prizes are better than fewer prizes and therefore choose the sure Integrating Individual Differences in Affect and Experience With Explanatory Theory Although the reactive route to risk taking highlights environmental factors, such as negative peer influences and other sources of temptation, reactions depend in part on the characteristics of the individual (Breiner, Stritzke, Lang, 1999). Caffray and Schneider (2000), for example, identify affective or emotional motivators that (a) promote risky behaviors by enhancing pleasant affective states, as in sensation seeking; (b) promote risky behaviors by reducing negative affective states, such as tension or depression; or (c) deter risky behaviors by avoiding anticipated regret. Consistent with these predictions, they found that adolescents who had had more experience with risky behaviors believed that those behaviors enhanced positive affect and reduced negative affect. Adolescents with less experience taking risks were more motivated to avoid negative future consequences. Cooper, Agocha, and Sheldon (2000) similarly found that adolescents with negative affect and avoidant personalities were more likely to engage in substance use and other risky behaviors, presumably to assuage their negative affect (see also Chassin, Pillow, Curran, Molina, & 20 Volume 7 Number 1